Research On Link Prediction Of Heterogeneous Information Networks Based On Frequent Subgraph Evolution

In recent years,with the popularization and development of information network technology,the amount of data generated in the network is growing rapidly,and the information contained in it is becoming richer and richer.It has become an inevitable trend to use the research of information network to guide people’s real life.As one of the main research contents of information network,link prediction aims to predict the possibility of links between nodes in information network.Link prediction has theoretical research value and is widely used in social networks,biomedicine,finance and other fields.The core problem of link prediction is to explore the evolution law and structural characteristics of the network.Although there are many link prediction methods at present,there are few methods to integrate multiple features.Most methods only consider topological features,semantic features or temporal features,and fail to effectively use feature fusion for link prediction.At the same time,most of the existing link prediction methods are based on homogeneous networks,while a large number of information networks have heterogeneous characteristics in real life.The research on heterogeneous information networks still needs to be deepened.To solve the above problems,this thesis proposes Link Prediction Model based on Community Division and Matrix Sliding（LP-CDMS）.Compared with the traditional methods,we fully consider the important micro network structure of frequent subgraph and explore its evolution process in the network.At the same time,the topological feature of community structure is used,and the semantic feature and topological feature are integrated.The specific research contents of this thesis are as follows:（1）Frequent Subgraph Mining Model based on Sequential Characteristics（SCFSM）is proposed.Considering the micro structure feature of frequent subgraphs in complex information networks,the frequent subgraph mining algorithm is used to mine the features after the graph division of complex information networks;Then the mining results are analyzed by time series model.（2）An improved FN（Fast Newman）algorithm is proposed.The algorithm divides a heterogeneous information network into multiple isomorphic information networks and multiple heterogeneous information networks,calculates the modularity of each network respectively,sums it to obtain the total modularity and normalizes it.The total modularity is continuously updated by FN algorithm to obtain the community division result when the modularity is the largest.（3）Link Prediction Model based on Community Division and Matrix Sliding（LP-CDMS）is proposed.The model uses the community division results of frequent subgraphs mined by SCFSM and FN algorithm with improved modularity.Using matrix sliding,the approximate subgraphs of frequent subgraphs are sliding matched in each community,the edges formed by the evolution of approximate subgraphs to frequent subgraphs are found,the attribute similarity of nodes is fused,and the probability of forming edges is calculated.（4）The feasibility of the proposed LP-CDMS algorithm is compared with some experimental data sets.